Research on threaded fastener detection in retired power batteries based on CSW-YOLOv8

Threaded fasteners are the primary connectors used in power batteries. The rapid and accurate detection of threaded fasteners is a critical step toward enabling human-robot collaboration and the automated disassembly of power batteries. In this study, a deep learning-based detection network, CSW-YOLOv8, was proposed based on the YOLOv8 architecture. This detection network was specifically designed for threaded fastener detection in power battery applications. To enhance shallow semantic representation and improve gradient propagation, a custom C3SDR feature extraction module was constructed and integrated into the backbone network. Additionally, a spatial and channel synergistic attention (SCSA) mechanism was introduced in the neck to alleviate the semantic deviation caused by diverse fastener shapes, occlusion, and complex backgrounds. To address the challenges of detecting small objects in cluttered environments, a novel Wise-EIoU loss function was designed to optimize bounding box regression. To validate the effectiveness of the proposed approach, a dataset consisting of 4450 images of threaded fasteners in power batteries was collected and augmented through various preprocessing techniques. The improved CSW-YOLOv8 model was trained and evaluated on this dataset, and all indicators reached a high level, which met the requirements for automated battery disassembly.